How we check claims.
Clear sources. Clear limits.

We start with the original source

We use the source that made the data. We tell you who the number describes, where it applies, and when it was collected.

We keep the source and the limits next to each number.

Statistics should link to the agency, university, research publisher, or study sponsor that made the data. A news story can add context, but it does not replace an available original table, report, questionnaire, or dataset.

Every number must keep the group counted, place, dates, definition, and uncertainty note.

A record, a survey, and a model are different

We label whether a number comes from official records, survey answers, or a calculation that combines separate sources. These are different kinds of evidence. We do not turn a survey answer into sales data or a population estimate into a count of active dating-app users.

We say one thing caused another only when the study can show it. Otherwise, we describe patterns and reported differences.

Who writes and checks the work

Articles name The Pool Report Research Desk and link to an authorship page. We do not make up a byline, credential, test, interview, or review.

Automation and AI can help with data work, drafts, code, checks, or design. Our standard stays the same: open the sources, check each claim, test the calculations, and state the limits.

Update and correction rules

We change a modified date only for an important update to facts, methods, structure, or sources. Small style edits do not make old evidence current.

We correct important errors quickly. We log them under the Corrections Policy. An update must not quietly erase a conclusion that changed in an important way.

Money does not change results

Advertisers, affiliates, platforms, and survey partners cannot buy a conclusion or a favorable comparison. They cannot hide a limit or an undisclosed link. We label paid relationships where readers see them.

The site does not load an advertising network. If that changes, we will update this policy and the related privacy notices before launch.

Treat people with care

The project can compare groups without calling a gender delusional. A physical or economic trait does not measure human value. Safety articles must separate warning patterns from claims about a person.

Do not use a result for employment, housing, credit, insurance, health care, or another important eligibility decision.